How a Better Dataset Creates a New SOTA Model!

Author(s): Boris Meinardus

Originally published on Towards AI.

Sometimes, it is enough to clean up the messy world of Multi-Modal AI Datasets to achieve a new SOTA model. We’ll look at the new MMICL paper: MMICL: Empowering vision-language model with multi-modal in-context learning [1] by researchers from China and the University of Washington.

Instead of focusing on simple image-to-text tasks, such as image captioning or visual question answering, this paper wants to design a model that performs very strongly in more complex and real-world multi-modal scenarios with interleaved images and text.

Examples of vision-language dialogue generated by MMICL. Source: [1]

case (a) for example, demonstrates how a user is asking the… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI